PHM-Net: A Domain-Prior-Guided High-Precision Measurement Network for Color Quantum Dot Optical Encoders

Abstract

As core positioning components in high-end precision equipment, optical encoders have been widely employed in high-end precision instruments and equipment. As an emerging measurement device, the quantum dot (QD) optical encoder can overcome the measurement precision limitations inherent in traditional black-and-white optical encoders. However, they suffer from complex degradations caused by QD diffusion, decay, and environmental illumination changes. In this study, we propose a novel high-precision measurement network, PHM-Net, which involves a red-guided color space reconstruction (R-g CSR) module, a U-shaped network, and a displacement measurement module. Specifically, to alleviate the influences of color distortion and noise contamination, a novel color space named RGB-Mr is constructed based on color priors of QDs to enhance the color characteristics of code pattern images. To weaken the influence of geometric distortion, an axial gated attention mechanism (AGAM) with dynamic attention is proposed to enhance the shape features of code patterns. To reduce the potential movement deviations of acquired code patterns, a composite loss with consistency constraints is proposed to guide the network optimization. Finally, a centroid-shift tracking algorithm is proposed to perform displacement measurement. The experimental results demonstrate that the proposed PHM-Net outperforms existing optical encoder measurement methods and state-of-the-art deep learning models, with improvements of at least 54.4 % in root-mean-square error (RMSE), 50.5 % in repeatability error (RE), and 67.9 % in variance.